Key Takeaway

Uncertainty around the application of privilege to AI Tools means that, for now, if you would not write it in an email that could be subpoenaed, do not type it into a consumer AI platform. Courts are coming to differing results regarding the protections available to generative AI output, and the risks remain serious.

Introduction

Much attention is being given to a February 17, 2026 ruling in United States v. Heppner, 25-cr-503, a federal judge in the Southern District of New York issued what the opinion itself (somewhat incorrectly) describes as a first-of-its-kind ruling on whether communications between a user and a generative artificial intelligence platform are protected by attorney-client privilege or the work product doctrine. The court held that neither protection applies.

Days before this, however, in Warner v. Gilbarco, Inc., 2:24-cv-12333, a judge in the Eastern District of Michigan ruled in a very different manner, providing extensive citation and discussion regarding why the use of AI as a litigation-related tool by a party (in this case, a pro se litigant) is protected by the work product doctrine.

While these rulings apply to federal law rather than Texas law, they each carry significant potential implications for businesses and individuals across the country who use AI tools in their daily routines. This alert summarizes the rulings, identifies some key limitations, and provides practical guidance for managing the shifting ground of AI-related risks in legal matters.

Heppner Orders Production

After learning he was the target of a federal investigation, but before retaining counsel, Heppner used a commercially available AI platform (Claude) to prepare documents "that outlined defense strategy" and "what he might argue with respect to the facts and the law."

He then retained counsel and emailed 31 documents that contained the input and output he created with Claude to his lawyers. His lawyers subsequently identified these documents to the government as part of a privilege protocol related to electronic discovery in the case, and the government sought production of the documents via a motion to compel.

Notably, the Heppner docket does not indicate that his counsel filed a response to the government’s motion, and the opinion states the issue was argued at a pretrial hearing. This indicates that the issue may not have been fully briefed, potentially leading to this result.

Attorney Client Privilege Does Not Apply

The court held that the AI-generated documents lacked at least two, if not all three, elements of the attorney-client privilege. This is not terribly surprising

First, AI is not an attorney. The court noted that [i]n the absence of an attorney-client relationship, the discussion of legal issues between two non-attorneys is not protected by attorney-client privilege."  Second, the communications were not confidential. Anthropic's privacy policy, to which Claude users consent, provides that the company collects data on both user "inputs" and Claude's "outputs," uses such data to "train" Claude, and reserves the right to disclose such data to "third parties." Third, Heppner did not communicate with Claude for the purpose of obtaining legal advice. Claude explicitly disclaims providing legal advice, stating when asked: "I'm not a lawyer and can't provide formal legal advice."

Work Product Doctrine Does Not Apply

In the more interesting ruling from Heppner, the court also rejected the application of work product protection, largely based on a finding that the documents were not "prepared by or at the behest of counsel" and did not reflect defense counsel's strategy. Because Heppner "acted on his own" when creating the documents, he was not acting as his counsel's agent, and the purpose of the doctrine—to protect lawyers' mental processes—was not implicated. The Court also discusses scenarios that would have potentially been a closer call, such as if defense counsel had directed Heppner to conduct these analyses but ultimately returns to what it finds to be outcome determinative – Heppner’s decision to undertake these actions on his own.   

Heppner - Limitations and Considerations

Several aspects of the Heppner decision warrant careful consideration.

The Heppner court applied federal case law and did not cite any procedural or evidentiary rules. Under Texas Rule of Civil Procedure 192.5, however, work product protection extends to "material prepared or mental impressions developed in anticipation of litigation or for trial by or for a party"—not solely material prepared by or for an attorney. Similarly, Federal Rule of Civil Procedure 26(b)(3) protects material prepared "by or for another party or its representative". The extent to which this scope was properly presented to the court in Heppner is unclear. Notably, the Southern District of New York’s own cases have held that a criminal defendant’s work product is protected, so it is difficult to square Heppner with this precedent. See, e.g., Wultz v. Bank of China Ltd., 304 F.R.D. 384 (S.D.N.Y. 2015) (holding that Federal Rule of Civil Procedure 26(b)(3)(A) accords the protection to material prepared “by or for [a] party or its representative”—not merely material prepared by or for an attorney); see U.S. v. Stewart, 287 F. Supp. 2d 461, 465-466 (S.D.N.Y. Oct. 20, 2003) (relying on the language of Federal Rule 26(b)(3) to protect a document prepared “by…a party”).

Additionally, Claude’s terms of service nuances were not fully addressed. The court's analysis did not address what would happen if a user disabled Claude's "train model" feature, the distinctions between free and paid versions of the platform, or the specific scope of Anthropic's data-sharing provisions. These factors could potentially affect the confidentiality analysis in future cases.

In the absence of other rulings, the Heppner decision has a chance to influence other courts who might feel uncertain about tackling this issue. Notwithstanding its limitations, courts facing similar requests may rely on this decision as persuasive authority until more developed case law gets similar attention.

Enter Warner v. Gilbarco, Inc.

In a ruling actually issued 7 days prior to Heppner, the Court in Warner addressed a motion to compel that specifically sought the production of all documents and information concerning a pro se plaintiff’s use of AI in connection with the lawsuit and to expressly overrule her claims of work product doctrine protection over them. The plaintiff had testified during her deposition that she used ChatGPT to ask legal questions related to the case, prepare documents, and upload materials for analysis. She resisted production of these documents as work product. Presuming the plaintiff used AI to assist in preparing her argument, she did so well, as the Court agreed with her.

To be clear, while the plaintiff in Warner was acting pro se, the court did not rely on that fact to support a finding that the work product doctrine applied. Additionally, the court’s logic and ruling would have been the same regardless of whether the AI tools were used in anticipation of litigation or during the case. To reach this conclusion, the court turned to the specific language of the Federal Rules of Civil Procedure to determine that the requested information was “not discoverable.” The court cited directly to Rule 26(b)(3)(A) for the point that a party may not discover documents “prepared in anticipation of litigation … by another party or its representative.”

The Warner court doubled down on this holding by spending the next two pages of its opinion citing a litany of cases for the proposition that even if the material was discoverable, it would be subject to protection under the work-product doctrine for a variety of reasons.  

The Court addressed head-on the contention that the use of an AI tool operates to waive protection by pointing out that a waiver of protection “has to be a waiver to an adversary or in a way likely to get in an adversary’s hand,” focusing on the substantive differences in waiver of attorney-client privilege and work product protection. In re Columbia/HCA Healthcare Corp. Billing Pracs. Litig., 293 F.3d 289, 306 n.28 (6th Cir. 2002); see also United States v. Am Tel. & Tel. Co., 642 F.2d 1285, 1299 (D.C. Cir. 1980) (“while the mere showing of a voluntary disclosure to a third person will generally suffice to show waiver of the attorney-client privilege, it should not suffice in itself for waiver of the work product privilege.”). This difference was not raised or analyzed in Heppner.

Finally, the court noted that ChatGPT and other generative AI programs are “tools, not persons, even if they have administrators somewhere in the background.” The court agreed with the plaintiff’s position that ChatGPT was simply a tool for a party to explore their own “internal analysis and mental impressions—i.e., her thought process” and forced production of these materials would clearly run afoul of the “heightened protection afforded to opinion work product."

So two different cases, briefed and analyzed very different – with starkly different results. One defendant having to turn over his core thought processes about his potential guilt, another plaintiff having “heightened” protection over hers. What are we to do make of this?

Implications for Texas-Based Businesses

• Texas work product rules may offer broader protection. Texas Rule of Civil Procedure 192.5 protects material prepared "by or for a party"—not just by counsel—mirroring the rules cited in the Warner case. This could support arguments for protecting AI-generated documents that would fail under the narrower standard applied in Heppner. But as we see in these two decisions, outcomes are uncertain.
• Unclear whether Federal courts in Texas will follow Heppner or Warner. Until Texas state courts or the Fifth Circuit address this issue, federal courts in Texas may treat either ruling as persuasive authority.
• Consumer AI use is not confidential in the traditional sense. Regardless of forum, Texas businesses should assume that communications with consumer AI platforms like ChatGPT and Claude are not confidential and that fact may be relied on by some courts analyzing their discoverability.
• Litigation holds need immediate review. Companies should update their litigation hold protocols and document retention policies to capture AI chat logs and prompt histories before disputes arise out of an abundance of caution.
• Enterprise AI reduces—but does not eliminate—risk. Deploying AI under negotiated enterprise agreements may improve confidentiality arguments, but attorney involvement remains the touchstone of establishing privilege.

Conclusion

Both Warner and Heppner decisions underscore that longstanding principles governing privilege and work product apply to new technologies, including generative AI. While questions remain about each ruling's precise scope—particularly under Texas law—the safest course is to treat consumer AI platforms as non-confidential channels unsuitable for sensitive legal communications while the courts begin to more fully analyze the application of privilege and work product to a rapidly evolving landscape of AI tools. If you have questions about how this ruling may affect your business or pending legal matters, please contact us.

About Gray Reed

Gray Reed is a full-service Texas law firm with a comprehensive range of litigation and transactional practices. Gray Reed’s more than 160 attorneys focus on understanding our clients’ businesses and industries, supporting companies and high-net-worth individuals with sophisticated legal needs. Through our subsidiary, Gray Reed Advisory Services, we extend our support with expert consultants who provide strategic business solutions tailored to each client’s objectives. Learn more at www.grayreed.com and www.grayreedadvisory.com.